Google's General Purpose AI, AlphaGo Explained!

What is AlphaGo?

AlphaGo is a General purpose AI by Google’s DeepMind. DeepMind is a British Company founded in Sep 2010 which was acquired by Google in 2014. AlphaGo is a Computer Program that plays Board Games. The latest Version of AlphaGo is AlphaGo Zero which is the one of the most intelligent AI in recent times. AlphaGo has defeated 18 times world champion Lee Sedol in a Chinese Classical Board Game GO. This level of Intelligence was not expected from machines at this stage of time. It is a general purpose AI because it’s not specific but does a whole General work by learning from its own experience. 

How does it work?

To start with, How do we play Tic-tac-toe, obviously we don’t analyze all the moves before making a move. We play with our intuition at any instant and so does AlphaGo. AlphaGo uses its intuition by seeing around an area where it’s opponent has last played. But not all AI programs do that. Previous AI’s play Games like chess and Tic-tac-toe by using Brute force techniques i.e they analyze all possible moves before making a move. A Classic Example is IBM Deep Blue Computer which was in the news for defeating a Champion Garry Kasparov in chess. But it was possible to analyze all moves in chess because not more than 20,000 combinations were possible at any time. But Considering Go, there are more than 2^100 moves possible on a 19x19 board game which is more than the total number of atoms present in the entire universe. So, practically it is not possible to analyze all moves before making one. So, it uses Reinforcement Learning and Neural science to mimic the human brain. Deepmind says that AlphaGo learns from its own experience without human intervention.

At the start, people at deep mind downloaded 1000’s of computer games played by amateur players and started to feed the AlphaGo program using pixels as dataInputs. Definitely, they wanted it to perform better than human. So after feeding, they made it play with itself 30 million times. And likewise, a newer version was made to play with its older version and thus improving itself on its own. The Program initially played some basic games and improved itself on its own to play many other difficult games.

What is the Goal of Deepmind?

The Goal is to solve intelligence by using best techniques from machine learning and neuroscience to generate more powerful general-purpose learning algorithms. They want to formalize intelligence. Find out what is Intelligence and how does it work. Our human brain is the best model to study intelligence and mimicking them in machines. However, the end goal is not just embedding intelligence into machines but to learn it’s working. 

Should we be Worried?

As an optimist, we should explore this science. As Newtonian mechanics has laid the groundwork for rockets and many other inventions. Deep learning has a world of its own. This could bring something very big in the near future considering its recent improvements and the pace at which it is improving. But, the only thing we should be worried is about the employability of humans, these machines are improving at such a great pace that they will soon beat humans in every possible field.

I hope this information helps. Finally, Thank you.